Background: When practices are randomized in a trial and observations are made on the patients to assess the relative effectiveness of the different interventions, sample size calculations need to estimate the number of practices required, not just the total number of patients.
Objective: Our aims were to introduce the methodology for appropriate sample size calculation and discuss the implications for power.
Method: A worked example from general practice is used.
Discussion: Designs which randomize practices are less powerful than designs which randomize patients to intervention groups, particularly where a large number of patients is recruited from each practice. Studies which randomize few practices should be avoided if possible, as the loss of power is considerable and simple randomization may not ensure comparability of intervention groups.